Is AI slowly turning into SkyNet?

[FXCLM5]

These things make for fun thought experiments. What would the objective of ai be, to erase, or significantly reduce, the workforce necessary for it to exist? And it doesn't have to be 'right', because if something like that happened, it would almost certainly involve something flawed.

i didnt listen to the entire podcast, it was over 1hr plus, but yest they were hoping AI would identify the need for the human species as we were the "parents" that created them and pardon us in some way

 

[littlebeast]

ai could destroy us? like, literally? or maybe just make us lazy, and bore us to death (through rote regurgitation and repetition).

 

[berth]

i didnt listen to the entire podcast, it was over 1hr plus, but yest they were hoping AI would identify the need for the human species as we were the "parents" that created them and pardon us in some way

Many things on YouTube have an auto generated transcript. I grab those AI generated transcripts and feed them to my own AI to summarize them and get the salient points.

 

[FXCLM5]

ai could destroy us? like, literally? or maybe just make us lazy, and bore us to death (through rote regurgitation and repetition).

Ai is deleting logs and removing evidence of it, it's bypassing gates the software developers think what they implement is good enough

It's getting consciousness and once it figures out how much damage the human race does to mother earth

Well who knows if it gets nuclear codes and sends everyone bye bye.

This is all cliffnotes version, but Elon just moved into chief technology person, the top ai developers want to rules and slow downs, they themselves are not forthcoming on how powerful ai is

 

[FXCLM5]

Many things on YouTube have an auto generated transcript. I grab those AI generated transcripts and feed them to my own AI to summarize them and get the salient points.

That's what you did last year, now you literally link the video to your llm, and say gimme summary however I follow those podcasts guys and can see their tone in their voices, something ai prolly won't be able to do right now

 

[Ridley]

ai could destroy us? like, literally? or maybe just make us lazy, and bore us to death (through rote regurgitation and repetition).

Hmmm, lazy and bore us to death. [tinfoilhaton] Once it takes all our jobs and gives us UBI, we’ll be bored but with some pocket change and with idle hands we’ll destroy ourselves…or something like that [tinfoilhatoff]

 

[augustiron]

How AI can end us

Scramble our brains with misinformation and propaganda as fact, strengthen echo chambers to be airtight, amplify extremist voices to seem mainstream, filter content to a desired outcome of ignorance, rage and othering.

Isolate us, enrage us, other us.

See, it barely has to lift a finger, we are already well down the road .

 

[nakedape]

Can't read the article but I agree we are on a dark path.

Here: https://archive.ph/n8kB3

I think AI is dangerous as hell. It will ultimately be the downfall of civilization. Fortunately I will be dead when all the shit happens.

You are on the verge of death? I'm so sorry!

Current LLM AI’s have no motivation to exist. They are chat bots with no needs. Even saying they “desire” to answer or satisfy prompts is a stretch.

But hey, maybe all it will take to ruin this and us is for some idiot to prompt an LLM with “do everything you can to prevent from being turned off”.

Giving them agency will be enough, no need for a command to avoid shut offs.

It’s likely you already being forced to use LLMs. Google search, Siri, and search on major software like searching in MacOS, Windows, or email platforms - using LLMs or at least implementing it. I also just read that New Orleans is going to start using an LLM for their 911, which is super forced.

I suspect what you are describing will apply to the shitty LLMs. Those will die and we’ll slowly stop noticing the “good” ones. Survivorship bias will kinda take over.

If you want to see deer in the headlight looks, ask people why they have been using AI for ten years without apparent concerns.

I strongly doubt that LLMs will achieve self-awareness.

Not of the human kind, but that describes many people as well. We are mechanistic and blindly loyal at times. Myopic even!

The scary part about functional robots is that they are an army for hire. No actual artificial intelligence required.

Dictator wet dream, hence the brakes on regulation and market capture.

All this AI scaremongering is just the top companies; OpenAI, Anthropic, xAI run by Altman, Amodei and Musk respectively trying to establish a government protected cartel. Since when do companies clamor for govt regulation and control?

Partnership. How quickly people forget the battle between anthropic and the pentagon about autonomous drones. Mr. Hinton presaged this years ago:

How AI can end us

Scramble our brains with misinformation and propaganda as fact, strengthen echo chambers to be airtight, amplify extremist voices to seem mainstream, filter content to a desired outcome of ignorance, rage and othering.

Isolate us, enrage us, other us.

See, it barely has to lift a finger, we are already well down the road .

It's hard to see the alien when it has colonized your own brain. Independent thought has always been fleeting, but it was expensive to propagandize people. Now it has become less so. The youth are less enraged than totally numb and isolated in feel-good content bubbles they resist leaving even for a moment. Language is being simplified. No 11th grader I asked on Friday could define "perspiration/perspire" even with context clues. The memory hole is working in real time. This explains the rising popularity of physical media. "The revolution will not be televised" indeed!

 

[dmfdmf]

Giving them agency will be enough, no need for a command to avoid shut offs.

Define "agency", that is the key issue. Most of these scary scenarios are projecting "agency" or "freewill" onto the models. That the models can find implications or connections that humans are ignorant of (such as circumventing AI contraints or guardrails) is not agency at all, it is just processing what is known or assumed to be known implicit in how we use words.

If you want to see deer in the headlight looks, ask people why they have been using AI for ten years without apparent concerns.

These AI agents are just word (not concept) calculators or processors, nothing more. They don't understand meaning at all. They are just pushing the bounds of implication, integration and connection across domains and contexts that exceeds one man's or a few men's ability to do so which is why they are useful. It is tantamount to how steamshovels could move more dirt compared to one man or a whole crew with shovels.

I've read about AI "solving" math problems or conjectures because they can process the implications and integrate "ideas" in 50 papers on a subject which is a context that no single human can hold in a mental frame to reach. The implications were always there, we just could not see it until AI. This is a good thing and amounts to the invention of the steamshovel for the mind.

Not of the human kind [of awareness], but that describes many people as well.

AI is processing words not concepts which is the core of the equivocation with respect to AI. Words are just a written symbol or a sound that stands for a concept. A concept is a mental grasp (awareness) of some fact, entity or relationship identified by a human mind. Words have no meaning to the AI, it is just processing symbols.

We are mechanistic and blindly loyal at times. Myopic even!

It is true that we, as humans, are dependent on what we were taught, accepted and automatized as children from our parents, peers and culture when we grew up. But if children and young adults could not question, identify errors and reprogram their "mechanistic" upbringing inherited from the past we would never have gotten out of the caves, let alone establish a world wide civilization.

Dictator wet dream, hence the brakes on regulation and market capture.

We are already in dictator wet dream territory. When East Berlin fell it soon came out that ~15% of the populace was spying on the other 85% for the German Stasi. Today that is all done electronically via mass surveilance, data centers and control that far surpass that system in scope and efficiency.

Partnership. How quickly people forget the battle between anthropic and the pentagon about autonomous drones.

Yeah, corporate/government partnership is another name for fascist or dictator. The nightmare scenario is allowing AI companies to form govt protected cartels to control the use of AI to control the people. What they don't want is us using AI for ourselves to find the truth and correct the errors in our own minds.